Indonesi
an
Journa
l
of El
ect
ri
cal
Engineer
ing
an
d
Comp
ut
er
Scie
nce
Vo
l.
12
,
No.
3
,
Decem
ber
201
8
, p
p.
968
~
973
IS
S
N: 25
02
-
4752, DO
I: 10
.11
591/ijeecs
.v1
2
.i
3
.pp
968
-
973
968
Journ
al h
om
e
page
:
http:
//
ia
es
core.c
om/j
ourn
als/i
ndex.
ph
p/ij
eecs
Classific
atio
n
of
Prost
ate Canc
er
usin
g Wavel
et N
eural
Network
Mohana
d
Naj
m A
b
dulw
ah
e
d
Mate
ri
al
s Dep
artm
ent
,
Univ
ersity of
T
ec
hno
log
y
,
Baghda
d,
Ira
q
Art
ic
le
In
f
o
ABSTR
A
CT
Art
ic
le
history:
Re
cei
ved
J
un
17
, 2
01
8
Re
vised
Ju
l
27
,
201
8
Accepte
d
Aug
2
0
,
201
8
Pros
ta
te
ca
nc
er
is
the
ce
n
tur
y
d
isea
se
tha
t
end
a
nger
th
e
l
ife
of
m
en.
Th
e
ea
rl
ie
r
to
d
ia
gno
se
the
dise
ase
,
th
e
proba
bi
li
t
y
of curing
thi
s di
se
a
se
is hi
gher
.
The
ref
or
e,
n
ew
appr
oac
h
es
of
di
agnosis
is
req
uir
ed
to
eff
ectivel
y
det
e
ct
th
e
prostat
e
ca
nc
er
i
n
ea
rl
y
st
age
co
m
par
ed
to
the
tr
adi
ti
on
al
m
et
hod
s.
The
ref
or
e
,
W
NN
is
a
new
adopt
ed
appr
oac
h
in
prost
ate
ca
n
ce
r
d
ia
gn
osis.
Morlet
func
ti
on
is
used
as
an
a
ct
iv
at
ion
func
ti
on
of
wav
el
e
t
neur
al
n
et
w
ork
(W
NN)
and
bac
k
prop
a
gat
ion
(BP)
is
a
ppli
ed
to
tra
in
t
he
W
ave
let
netw
ork.
W
NN
cl
assifi
es
prostate
ca
n
ce
r
a
cc
ordi
ng
to
thre
e
f
ac
t
ors:
pat
ie
n
t
age,
PS
A
le
vel,
and
prostat
e
v
olume.
W
NN
per
form
anc
e
is
eva
luated
bas
ed
on
the
per
ce
n
ta
g
e
of
c
la
ss
ifi
c
at
ion
and
the
comput
at
io
nal
complex
ity
of
seve
ra
l
ca
ses.
The
resul
t
s
o
f
the
sim
ula
t
i
on
show
tha
t
W
NN
has
lower
m
ea
n
squar
ed
err
or
(MS
E) than the Neur
a
l
Ne
t
work (NN).
Ke
yw
or
d
s
:
Ar
ti
fici
al
n
e
ur
a
l netw
ork
Pr
ost
at
e cance
r
Wav
el
et
ne
ur
al
n
et
w
ork
Copyright
©
201
8
Instit
ut
e
o
f Ad
vanc
ed
Engi
n
ee
r
ing
and
S
cienc
e
.
Al
l
rights re
s
erv
ed
.
Corres
pond
in
g
Aut
h
or
:
Moh
a
na
d Najm
A
bd
ul
wah
e
d
,
Ma
te
rial
s D
epa
rtm
ent,
Un
i
ver
sit
y o
f Te
ch
no
l
og
y,
Ba
ghda
d,
Ir
a
q
.
Em
a
il
:
m
oh
anad
.
naj
m
@yahoo
.co
m
1.
INTROD
U
CTION
To
day,
prostat
e
cancer
ha
s
be
com
e
on
e
of
the
m
os
t
dan
ge
rous
disease
s
that
can
af
fect
the
healt
h
of
m
en,
a
nd
the
de
te
ct
ion
of
the
disease
i
n
ea
rly
sta
ges
ca
n
he
lp
to
get
rid
of
the
disea
se
e
asi
ly
.
The
t
rad
i
ti
on
al
m
et
ho
ds
f
or
di
agnosis
the
prostat
e
cancer
de
pende
d
on
in
div
id
ual
bi
om
a
rk
e
rs
an
d
te
ste
d
the
ti
ssu
e
an
d
cel
l
sp
eci
m
ens.
The ge
netic
and
a
ge
are
the c
oeffici
ents that
he
lp to diag
nosis
the prost
at
e cancer
, and
t
hey
are the
first
data
m
us
t
be
avail
able
to
determ
ine
the
disease.
Pr
ost
at
e
-
S
pecif
ic
An
ti
gen
(PSA)
is
the
pr
otei
n
bio
m
ark
er
that
us
ually
us
e
d
to
detect
the
pa
ti
ent
of
pro
sta
te
cance
r
[
1]
.
T
he
ra
ng
e
of
t
he
PSA
is
rec
ord
ed
lo
w
(less
than
t
wo
ng
/m
L)
in
the
blood
of
t
he
he
al
thy
m
en,
an
d
it
is
raised
f
or
the
prostat
e
cancer
patie
nts
m
or
e
th
an
t
wo
[
2]
.
Althou
gh
the
P
SA
is
a
n
im
po
r
ta
nt
ind
ic
at
or
t
hat
can
detect
t
he
prostat
e
can
cer
disease
,
the
oth
e
r
m
ul
ti
ple f
act
ors sho
uld
be
ta
ke
n
int
o
acc
ount
to
the
dia
gnos
i
s of the
il
lness
su
c
h
as a
ge
a
nd
gen
et
ic
[3]
.
Ther
e
a
re
seve
ral
stud
ie
s
ha
ve
discuss
e
d
the
diagnosis
of
the
prostat
e
can
cer
in
early
stag
es
ap
p
ly
ing
m
or
e
recent
and
e
ff
ect
ive
m
et
hods
.
Kall
en
et
al
.
[4]
us
e
d
an
e
xam
ple
of
C
onvoluti
onal
Ne
ur
al
Ne
twork
(CN
N)
a
nd
he
ap
p
li
ed
bo
t
h
Suppor
t
Vecto
r
Ma
chi
ne
(
S
VM)
a
nd
a
pre
-
trai
ne
d
CN
N
for
cl
assifi
cat
ion
a
nd
featur
e
e
xtracti
on
res
pecti
vel
y.
I
n
a
uto
m
at
e
d
cl
assifi
cat
io
n
of
Glaso
n
gr
a
ding,
G
or
el
ik
et
al
.
[
5]
segm
ented
the
m
ic
ro
scop
i
c
i
m
ages
into
a
m
eaning
f
ul
pa
tho
lo
gical
ly
segm
ents
su
ch
as,
L
um
en,
sto
rm
a
and
lym
ph
ocyt
e
et
c.
D
oyle
et
al
.
[
6]
propo
sed
a
Ba
yse
ia
n
m
ulti
-
reso
lu
ti
on
m
et
ho
d
f
or
cl
assifi
cat
ion
usi
ng
Ad
a
Boo
s
t
.
In
To
rste
n
et
al
.
[
7]
us
e
d
a
cl
assifi
cat
ion
by
le
ar
ning
ve
ct
or
quantiz
at
ion
in
w
hich
the
Neural
Net
works
trai
n
in
g
we
re
adap
te
d
int
o
di
ff
e
ren
t
s
a
m
ple
siz
es
per
cl
ass
.
A
no
t
her
recent
researc
h
art
ic
le
by
Lit
j
ens
et
al.
[8]
us
in
g
C
NN
f
or h
em
at
ox
yl
in classi
ficat
ion
and eo
sin im
ages of prosta
ti
c ti
ssu
e.
Wav
el
et
netw
orks
we
re
fir
stl
y
introd
uced
by
Zha
ng
a
nd
Be
nveniste
[9]
as
a
n
al
te
rn
at
iv
e
a
ppr
oach
t
o
FeedFo
rw
a
rd
Neural
Net
wor
ks
(
FF
NN)
tha
t
fix
the
dr
a
w
ba
cks
of
NNs
a
nd
Wa
velet
A
naly
sis(
WA
)
wh
il
e
it
has
the b
est
p
e
rfor
m
ance u
sin
g
both of
these
ap
proac
hes
[
10]
.
Chen
, F
en
g
[10]
us
e
d Wa
velet
N
eu
ral N
et
work
Evaluation Warning : The document was created with Spire.PDF for Python.
Ind
on
esi
a
n
J
E
le
c Eng &
Co
m
p
Sci
IS
S
N:
25
02
-
4752
Cl
as
sif
ic
ation
of Pr
os
tate C
ancer
us
in
g Wa
vel
et
Neura
l N
et
work
(
Mo
hanad N
aj
m
A
bdul
wah
e
d
)
969
(
WN
N
)
in
sys
tem
m
od
el
li
ng
and
tim
e
series
pr
e
dicti
on
base
d
on
m
ult
iresolutio
n
le
a
rn
i
ng.
Their
r
esults
ind
ic
at
ed
that
WNN
ha
s
a
powe
rful
ap
pro
xim
a
ti
on
ca
pa
bili
ty
and
su
it
abili
ty
in
pr
ed
ic
ti
on
an
d
m
o
deling.
Ther
e
f
or
e,
WNN is a
sig
nifica
nt to
ol in
cl
as
sific
at
ion
a
nd si
gn
al
s
pro
ce
ssin
g
[11]
-
[
12]
.
In
t
his
arti
cl
e,
a
new
a
ppr
oac
h
f
or
cl
assi
ficat
ion
of
prostat
e
cancer
’s
ris
k
is
propose
d
usi
ng
WNN
.
The
sim
ulati
o
n
of
WNN
m
e
thod
is
pe
rform
ed
to
get
hi
gh
cl
assifi
ed
re
su
lt
s.
The
ne
w
appro
a
ch
sim
ulati
on
resu
lt
is
com
pa
red
with
the
tra
diti
on
al
cl
assif
ic
at
ion
appr
oa
ches
su
c
h
as
N
eur
al
Net
works
(NN)
.
This
art
ic
le
is
orga
nized
as
th
e
fo
ll
ows.
Part
2
introduc
es
the
data
colle
ct
ion.
Part
3
pres
ents
W
N
N
wit
h
a
descr
i
ption
of
it
s
app
li
ed
le
ar
nin
g
al
gorithm
.
In
Part
4
t
he
si
m
ulati
on
resu
lt
s
of
both
NN
a
nd
W
NN
is
su
m
m
arized.
Finall
y, Part
5 pro
vid
es t
he
c
oncl
us
i
on of thi
s p
a
per.
2.
DA
T
A COLL
ECTION
Find
i
ng
a
nd
co
ll
ect
ing
the
dat
a
is
ver
y
im
po
rtant
f
or
trai
nin
g
a
nd
te
sti
ng
the
WN
N
a
nd
NN.
I
n
this
work,
the
data
will
be
div
id
ed
into
th
ree
s
ect
ion
s,
Ag
e
,
PSA
a
nd
the
Pr
ost
at
e
vo
lu
m
e.
Fo
r
hum
a
n’
s
a
ge
,
the
risk
of
pr
os
ta
te
cancer
i
s
hig
hly
ex
pe
ct
ed
sta
rting
f
ro
m
the
age
of
50
a
nd
ab
ove.
PSA
is
a
protei
n
pro
du
ce
d
by
th
e
m
align
ant
an
d
the
norm
al
c
el
ls
of
the
pros
ta
te
gland
[13]
.
The
ra
ng
e
of
PSA
m
easur
e
m
ents
ranges
f
ro
m
2
to
4
(
ng
/m
l)
in
the
blood
f
or
the
healt
hy
m
e
n.
H
oweve
r,
th
e
in
dicat
ion
of
PSA
var
ie
s
wit
h
age
as
it
sh
ow
s
a
ver
y
hi
gh
pro
bab
il
it
y
of
ca
ncer
with
PS
A
value
m
or
e
than
4
ng
/m
l
and
a
n
age
over
50.
Howe
ver,
feat
ur
es
of
P
SA
diag
nosin
g
the
disease
decr
e
ases
with
pati
ents
of
P
r
os
ta
te
cancer
with
la
rge
vo
l
um
e
of
P
r
ost
at
e
beca
use
t
he
c
o
-
occ
urrin
g
decr
ea
ses
th
e
eff
e
ct
of
t
he
cancer
on
PS
A
[
14
]
.
The
vol
um
e
of
the
prostat
e
in
dicat
e
an
ap
parent
sy
m
pto
m
wh
e
n
it
has
e
nlarg
e
d
af
fecti
ng
the
ur
et
hr
a
.
T
hat
patie
nts
dat
a
hav
e
been p
r
ov
i
ded
by Bag
hdad
Medica
l C
it
y ho
s
pital
in
A
pri
l 2
017.
3.
WA
VELE
T N
EUR
AL
NETWOR
KS
(W
N
Ns)
Wav
el
et
A
nal
ysi
s
is
the
resu
lt
of
both
F
ourier
T
ra
n
sf
orm
(F
T)
an
d
W
i
ndow
F
ouri
er
Tra
ns
f
or
m
(
WFT)
to
im
pro
ve
the
sho
rtcom
ing
s
of
FT.
F
ourier
tr
ansfo
rm
is
a
com
m
on
m
et
h
od
in
dig
it
al
sign
a
l
processi
ng.
H
oweve
r,
wh
e
n
it
was
re
ported
i
n
the
tim
e
seri
es
pr
e
dicti
on,
it
sh
owed
the
ti
m
e
lim
i
ta
ti
on
s.
WA
is
an
analy
t
ic
al
and
m
at
he
m
a
ti
cal
too
l
fo
r
a
wide
ra
ng
e
of
researc
h
ap
plica
ti
on
s.
Wav
el
et
An
al
ysi
s
(W
A
)
is
us
e
d
pr
ese
ntly
for
ti
m
e
po
sit
ion,
a
naly
sis
of
tim
e
series
an
d
inte
ns
it
y
[
15]
The
wav
el
et
i
s
a
s
pecial
wa
ve
form
with
a
finite
durati
on
at
the
a
ver
a
ge
zer
o
po
int.
The
WA
de
plo
ys
a
uniq
ue
functi
on
i
ntr
oduce
d
as
the
m
oth
er
wa
velet
.
This
wav
el
et
un
i
que
functi
on
is
t
he
re
su
lt
of
a
s
eries
of
basic
or
t
hogonal
set
s
com
po
se
d
of
a
fathe
r
wav
el
et
a
nd a
m
oth
er w
a
velet
w
hic
h
sat
isfie
s
:
∫
(
)
=
1
(1)
∫
(
)
=
0
(2)
The
wav
el
et
fa
m
ily
has
w
hat
is
cal
le
d
a
s
the
childre
n
of
w
avelet
,
w
hich
i
s
a
translat
ed
and
dilat
ed
form
o
f wavele
t
m
oth
er:
(3)
4.
WA
VELE
T N
EUR
AL
NETWOR
K
CON
FIGUR
ATIO
N
(
WN
N
C)
The WN
N
con
fig
ur
at
io
n
i
s
presented
and
dis
play
ed
in
Fig
ur
e
1
.
T
he
sig
nifi
cant str
uctu
re
of
WN
N
is
introd
uced as
t
he follo
wing
[
16
]
:
1
.
WNN
Para
m
et
ers
init
ia
lizat
ion
T
he
m
at
rix
that
e
xis
ts
betwee
n
t
he
input
a
nd
the
hidden
la
ye
r
is
=
(
)
×
The
m
at
rix
tha
t exist
s b
et
wee
n
the
h
i
dd
e
n an
d
the
outp
ut la
ye
r:
=
(
)
×
Hidden
lay
er
ne
uro
n’
s
D
il
at
io
n vecto
r:
=
(
1
,
1
,
…
)
Hidden
lay
er
ne
uro
n’
s
Tr
a
ns
l
at
ion
vecto
r:
Wh
e
re
m
,
p,
n,
sta
nd
s
f
or
the
in
pu
t
la
ye
r,
hidden
la
ye
r
an
d
t
he
in
put
la
ye
r
of
WNN
resp
e
ct
ively
.
The param
et
ers
init
ia
li
zat
ion
is do
ne
ar
bitrar
il
y.
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2502
-
4752
Ind
on
esi
a
n
J
E
le
c Eng &
Co
m
p
Sci,
Vo
l.
12
, N
o.
3
,
Dece
m
ber
2
01
8
:
968
–
973
970
Figure
1.
W
a
ve
le
t Neural
N
et
work c
onfig
ur
a
ti
on
2
.
Ge
ner
at
e t
he
outp
ut of t
he n
et
work b
y c
ompu
ti
ng t
he fo
r
ward
pass by a
pp
ly
in
g
the
fol
lowing
E
quat
io
n:
,
(
∑
=
1
)
=
1
√
(
(
1
)
−
)
(4)
W
he
re
:
(
1
)
=
∑
=
1
(5)
The
it
h n
ode
outp
ut at the
out
pu
t l
ay
er
is e
xpresse
d
as:
(6)
wh
e
re
X
is
th
e
W
N
N
i
nput
vecto
r
an
d
it
is
ex
pr
esse
d
as
=
(
1
,
2
,
…
,
)
.
w
he
n
X
is
a
chieve
d,
t
he
jt
h
ou
t
pu
t
node
of
the h
i
dd
e
n
la
ye
r
ca
n be fo
und ou
t.
3
.
C
om
pu
te
the
outp
ut total
er
ror by ap
plyi
ng the
foll
owin
g eq
uation:
(7)
Wh
e
re
sta
nd
s
f
or
the
t
otal
error,
sta
nd
s
f
or
t
rainin
g
sam
ples
nu
m
ber
f
or
e
ach
sam
ple
q,
sta
nd
s
for
the
desire
vecto
r.
The
desire
ve
ct
or
is
ex
press
ed
as
=
(
1
,
2
,
…
,
)
,
an
d
stan
ds
f
or
th
e
out
pu
t
vecto
r
and
it
is
e
xpr
essed
by
=
(
1
,
2
,
…
,
)
.
T
o
be
a
ble
to
a
chieve
the
outpu
t
m
in
i
m
u
m
total
error
is
consi
der
e
d
a
si
gn
i
ficant a
dv
a
ntage o
f WN
N .
4
.
Fin
d
eac
h pa
ram
et
er p
arti
al
d
eri
vatives as
expresse
d by
,
,
and
(8)
(9)
(10)
Evaluation Warning : The document was created with Spire.PDF for Python.
Ind
on
esi
a
n
J
E
le
c Eng &
Co
m
p
Sci
IS
S
N:
25
02
-
4752
Cl
as
sif
ic
ation
of Pr
os
tate C
ancer
us
in
g Wa
vel
et
Neura
l N
et
work
(
Mo
hanad N
aj
m
A
bdul
wah
e
d
)
971
(11)
wh
e
re
5
.
De
fi
ne
t
he
l
earn
i
ng r
at
e a
nd m
o
m
entum
v
al
ues
as α=
0.9
9
a
nd η =
0.2 t
o upd
at
e
WNN
p
a
ram
et
ers
as
sh
ow
n
in
E
qu
a
ti
on
s
(12
-
15)
[
15, 16]
(12)
(13)
(14)
(15)
5.
RESU
LT
S
AND DI
SCUS
S
ION
This
pa
rt
disc
us
ses
WNN
in
dia
gnos
is
of
prostat
e
canc
er.
T
hr
ee
different
it
e
m
s
wer
e
us
e
d
to
m
easur
e
prosta
te
cancer
ris
k,
patie
nt
a
ge,
PS
A,
a
nd
prostat
e
volu
m
e.
In
thi
s
arti
cl
e,
WNN
is
ap
plied
as
a
ne
w
m
et
ho
d:
it
con
sist
s
of
a
n
input
la
ye
r
(m
),
hi
dd
e
n
la
ye
r
(p)
an
d
a
n
ou
t
pu
t
la
ye
r
(
n).
The
in
put
la
ye
r
,
ou
t
pu
t
la
ye
r
a
nd
the
hidden
la
ye
r
co
ns
ist
s
of
three
node
s,
on
e
node
a
nd
se
ve
n
nodes
res
pe
ct
ively
as
sho
wn
in
the foll
owin
g
e
qu
at
io
n [
16
]
.
ℎ
≥ (2
*
)
+
1
(16)
Wh
e
re
ℎ
a
nd
de
no
t
es
the
ne
uro
ns
num
ber
i
n
t
he
hidde
n
a
nd
the
in
pu
t
la
ye
r
res
pecti
vely
.
Morlet
f
un
ct
io
n
is
de
plo
ye
d
in
the
hidde
n
la
ye
r
as
an
act
iv
at
ion
f
unct
ion,
and
s
igm
oid
functi
on
is
de
pl
oye
d
in
ou
t
pu
t
la
ye
r
[11]
.
Eq
uation
17
belo
w
ind
ic
at
es
the
Morlet
functi
on,
as
dis
play
ed
in
Fig
ure
2.
Table
1
pr
ese
nts t
he
ini
dicat
ion
of th
a
pp
li
ed
p
a
pram
et
ers
in t
his
work.
(17)
Wh
e
re,
t i
s the
su
m
m
a
ti
on
out
pu
t
f
un
ct
io
n
i
n hid
den lay
er.
Figure
2.
W
a
ve
le
t M
or
el
t F
unct
ion
[
11]
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2502
-
4752
Ind
on
esi
a
n
J
E
le
c Eng &
Co
m
p
Sci,
Vo
l.
12
, N
o.
3
,
Dece
m
ber
2
01
8
:
968
–
973
972
Table
1
.
Param
et
ers
I
ndic
at
ion
No
.
Para
m
eter
’s Na
m
e
W
NN
NN
1
Nu
m
b
e
r
o
f
neu
ron
s I
n
p
u
t lay
e
r
3
3
2
Nu
m
b
er
o
f
neu
rons Hidd
en
lay
e
r
7
7
3
Nu
m
b
e
r
o
f
neu
ron
d
Outp
u
t lay
er
1
1
4
Actv
atio
n
f
u
n
ctio
n
in h
id
d
en
lay
er
Morlet
sig
m
o
id
5
Actv
atio
n
f
u
n
ctio
n
in o
u
tp
u
t lay
er
sig
m
o
id
tan
sig
6
Lear
n
in
g
r
ate
valu
e (
η
)
0
.2
0
.5
7
Mo
m
en
tu
m
valu
e
(
α)
0
.99
0
.99
8
iteration
nu
m
b
er
2
0
0
epo
ch
2
0
0
epo
ch
9
Tr
ain
in
g
m
eth
o
d
b
ackp
rop
o
g
atio
n
Each
net
wor
k
fo
rm
s
200
e
po
c
hs
based
on
the
data
el
e
m
ent
network
of
Ba
ck
pro
pogatio
n'
s
BP
al
gorithm
.
On
ce
the
trai
ning
is
ov
er
,
WNN
is
directl
y
app
li
ed
f
or
t
he
cl
assifi
cat
ion
of
prostat
e
canc
er
risk
.
WNN
pe
rfor
m
ance ca
n be est
i
m
at
ed
dep
e
nding o
n
t
he follo
wing in
dicat
or
s.
5
.
1.
Conv
er
ge
nce of Per
f
or
man
ces
Figures
3
an
d
4
sho
w
the
M
SE
of
WN
N
a
nd
N
N,
res
pec
ti
vely
.
Figure
3
in
dicat
e
s
that
WNN
has
a
lowe
r
MSE
th
an
NN.
It
is
obvi
ously
dem
on
strat
ed
that
W
N
N
has
a
lo
wer
MSE
valu
e
tha
n
neural
netw
ork
,
wh
ic
h
ind
ic
at
e
s
the
high
perf
or
m
ance
of
the
cl
assifi
cat
ion
,
wh
ic
h
m
eans
a
hig
h
perform
a
nce
in
the
dia
gnos
i
s
of pro
sta
te
ca
nc
er. As a
r
es
ult,
WNN
has
a
hi
gh
e
r
c
onverge
nce
rate c
om
par
ed
to N
N.
Figure
3
.
MSE
for
WNN
Figure
4. MSE
for
N
N
5
.
2
.
Clas
si
ficat
i
on
E
xp
eri
m
ent
al R
es
ults
Table
2
s
hows
that
the
cl
assifi
cat
ion
res
ults
of
WNN
outp
erfor
m
the
cl
as
sific
at
ion
res
ults
obta
ine
d
by
NN
base
d
on
se
ver
al
cas
es
of
P
ro
sta
te
cancer
patie
nts
.
Furtherm
or
e,
the
resu
lt
s
of
the
cases
with
ages
above
50 a
nd P
SA
over
4 n
g/
m
l i
nd
ic
at
e a hi
gh
e
r
cl
assifi
ca
ti
on
resu
lt
s
of
WNN c
om
par
e
d
to
NN.
Table
2
.
Cl
assi
ficat
ion
of
WNN a
nd NN
Cas
e
Ou
tp
u
t
NN.%
Ou
tp
u
t
W
NN
.
%
Vo
lu
m
e
(c
m
)
(PSA
)
Ag
e
Na
m
e
n
o
r
m
al
6
.6
3
.6
8*6
1
.06
60
Ca
se 1
n
o
r
m
al
4
5
.61
3
0
.61
10*7
0
.39
84
Cas
e 2
Ab
n
o
r
m
al
9
0
.69
9
3
.69
(7
-
15)
*
(10
-
13
9
5
.11
72
Cas
e 3
Ab
n
o
r
m
al
8
2
.6
9
2
.6
1
6
1
.9
60
Cas
e 4
Ab
n
o
r
m
al
9
3
.99
9
9
.99
1
1
.3
70
Cas
e 5
Ab
n
o
r
m
al
9
2
.69
9
9
.69
5
4
.9
85
Cas
e 6
Ab
n
o
r
m
al
2
4
.6
2
4
.6
7
.48
50
Cas
e 7
Ab
n
o
r
m
al
9
9
.99
9
9
.99
1
6
.2
75
Cas
e 8
Ab
n
o
r
m
al
9
6
.88
9
9
.88
1
4
.11
67
Cas
e 9
Ab
n
o
r
m
al
9
2
.99
9
9
.99
6
.9
67
Cas
e 10
Ab
n
o
r
m
al
9
0
.6
9
2
.6
100
76
Cas
e 11
Ab
n
o
r
m
al
9
0
.9
9
9
.9
8
.28
61
Cas
e 12
Evaluation Warning : The document was created with Spire.PDF for Python.
Ind
on
esi
a
n
J
E
le
c Eng &
Co
m
p
Sci
IS
S
N:
25
02
-
4752
Cl
as
sif
ic
ation
of Pr
os
tate C
ancer
us
in
g Wa
vel
et
Neura
l N
et
work
(
Mo
hanad N
aj
m
A
bdul
wah
e
d
)
973
5
.
3
.
Co
m
pu
t
at
i
onal com
plexit
y
Com
pu
ta
ti
on
al
com
plexity
is
consi
der
e
d
a
s
a
sign
i
ficant
fac
tor
for
e
valuati
ng
the
pe
rfor
m
ance
of
the
netw
ork.
Ta
ble
3
il
lustrate
s,
the
C
PU
tim
e
util
iz
at
ion
i
n
the
sam
e
pr
oc
esso
r
durin
g
the
trai
ni
ng
for
ea
c
h
appr
oach.
T
he
instal
le
d
m
e
m
or
y
(RAM)
has
3G
B
us
in
g
I
ntel
Core
i3
M
CPU
37
0
2.40
G
Hz
proce
ss
or.
The
ti
m
e
uti
li
z
at
ion
res
ults
of
the
CPU
i
nd
ic
at
e
the
sp
ee
d
e
ff
ic
ie
ncy
of
W
NN
t
han
N
N.
Fu
rt
her
m
or
e,
it
has
been f
ound t
ha
t
W
N
N has
f
e
wer o
per
at
io
ns t
han NN m
et
ho
d. T
her
e
f
or
e,
WNN is
not as
co
m
plex
as
N
N.
Table
3
. CP
U uti
li
zat
ion
tim
e
Tr
ain
in
g
App
ro
ach
W
NN
NN
CPU ti
m
e(se
c)
1
0
.19
0
1
8
.22
6.
CONCLUS
I
ON
In
t
his
arti
cl
e,
WNN
is
us
ed
f
or
t
he
dia
gnos
i
s
of
prostat
e
ca
ncer.
T
he
Mo
rlet
functi
on
is
de
plo
ye
d
in
the
hidde
n
la
ye
r
as
an
act
ivat
ion
f
unct
io
n
w
hile
BP
m
et
hod
is
ap
plied
f
or
trai
ning
the
ne
twork
.
The
res
ults
of
the
sim
ulatio
n
sh
ow
that
WNN
pe
rfor
m
ance
works
m
uch
be
tt
er
than
neur
al
networ
ks
in
the
cl
assifi
cat
ion
of
prostat
e
cance
r
.
The
res
ults
s
how
t
hat
WNN
co
nver
ges
f
ast
er
tha
n
N
N,
an
d
the
c
onve
rg
e
nce
rate
of
MSE
is
slow
e
r
at
1.6
05e
-
10.
I
n
a
ddit
ion
,
WN
N
intr
oduces
bette
r
c
la
ssific
at
ion
va
lues
an
d
e
xhibi
ts
le
ss
com
pu
ta
ti
on
al
com
plexity
than
N
N.
H
ow
e
ve
r,
c
om
par
ed
t
o
N
N,
t
he
W
NN
nee
ds
m
ore
tim
e
to
achieve
the
best
va
lue
of
MSE beca
us
e
it
h
as a
lo
wer
le
arn
i
ng r
at
e t
o o
btain
higher
conv
e
r
gen
ce
r
at
e.
REFERE
NCE
S
[1]
W
olf,
A.,
e
t
al.,
Am
eri
ca
n
Cance
r
Socie
t
y
guid
el
i
ne
for
the
ea
rl
y
det
e
ct
ion
of
pro
stat
e
c
ancer:
up
dat
e
2010
.
CA:
a
ca
nc
er journa
l
fo
r
clinicians,
201
0.
60(2):
p.
70
-
9
8.
[2]
Van
Cangh,
P.J.
,
et
al.,
Free
to
tot
al
pros
tat
e
-
sp
ec
i
fi
c
an
ti
gen
(
PSA
)
ratio
improve
s
the
d
iscriminat
ion
be
twee
n
pros
tat
e
can
ce
r
and
beni
gn
pros
tat
ic
hyp
erplasia
(
BP
H)
in
the
d
i
agnostic
gray
zone
of
1.
8
to
10
n
g/mL
tot
al
PSA
.
Urolog
y
,
1996
.
48(6):
p.
67
-
70
.
[3]
Brawe
r,
M
.
K.,
Pros
ta
te
‐spec
ifi
c
antigen:
Curre
n
t
stat
us
.
CA:
a
c
anc
er
journ
al
fo
r
clinicians,
199
9.
49(5):
p.
264
-
281.
[4]
Käll
én
,
H.,
et
a
l.
Towa
rds
gra
ding
gle
ason
score
using
gene
rically
traine
d
de
ep
c
onvolut
ional
ne
ura
l
net
works
.
i
n
Bi
omedi
cal
Ima
ging
(
ISBI
),
201
6
IEEE
13
th
Int
e
rna
ti
on
al
S
y
m
po
sium
on.
2016.
I
EE
E
.
[5]
Gorel
ic
k
,
L.,
et
al
.
,
Pros
ta
te
hi
stopat
holog
y
:
L
ea
rning
t
issue
c
om
ponent
histogra
m
s
for
ca
nce
r
det
e
ction
an
d
cl
assifi
ca
t
ion
.
IE
EE
transacti
ons
on
medic
a
l imaging,
2013
.
32(10
):
p.
1804
-
1818.
[6]
Do
y
le,
S.
,
e
t
a
l
.
,
A
boosted
Ba
yesia
n
m
ult
ir
esol
uti
on
c
la
ss
ifi
er
f
or
prostat
e
ca
n
c
er
detec
ti
on
fro
m
digi
ti
z
ed
ne
ed
le
biopsie
s.
I
EEE
t
rans
act
ions o
n
b
iomedi
ca
l
eng
in
ee
ring,
2012.
59
(5):
p.
1205
-
121
8.
[7]
Matt
fe
ldt
,
T
.
,
et
al
.
,
Cl
assificat
io
n
of
prostatic
ca
r
ci
nom
a
with
arti
fi
cial
n
eur
al
ne
t
works
using
compara
ti
ve
geno
m
ic
h
y
brid
izati
on
an
d
quantitative st
ere
olog
ic
a
l
d
at
a
.
Pat
ho
logy
-
R
ese
arch
and
Pra
ct
i
c
e
,
2003
.
199
(12)
:
p.
773
-
784.
[8]
Li
tjens,
G.,
e
t
al.,
Dee
p
l
earning
as
a
tool
for
inc
reased
acc
urac
y
and
ef
ficie
n
cy
of
histopat
holo
gic
al
diagnosis
.
Scie
nti
f
ic re
por
t
s,
2016.
6:
p
.
26
286.
[9]
Zha
ng,
Q. and
A
.
Benv
eni
st
e, Wave
l
et
n
et
works
.
IEE
E
transacti
o
ns on
Neural
N
etwor
ks,
1992.
3(6
):
p.
889
-
898.
[10]
Chen,
Z.,
T.
Fe
ng,
and
Q.
Meng.
The
appli
cati
on
of
wave
le
t
n
e
ural
net
work
in
ti
me
series
pr
edi
ct
ion
and
syste
m
modeli
ng
based
on
multi
resolution
le
arning.
in
S
y
stems
,
Man,
a
nd
C
y
ber
n
etics,
1999.
IEEE
SM
C'
99
Confer
ence
Proce
edi
ngs.
19
99
IEEE
In
te
rna
t
iona
l
Confer
e
nc
e
on.
1999.
IE
E
E.
[11]
Ghawbar,
F.M.,
et
al.,
PA
T
an
d
Px
Code
Sid
e
lobe
R
educ
t
ion
Us
ing
W
ave
let
Neural
Net
work
,
in
Advanc
es
i
n
Mac
hine L
ea
rni
ng
and
Sign
al Proce
ss
ing.
2016
,
Springer.
p
.
117
-
128.
[12]
Ahm
ed,
M.S.,
N
.
S.M.
Shah,
and
S.I.
Anka,
Sidelobe
red
uction
using wa
vel
et
neu
r
al
net
work for
bina
r
y
cod
ed
pulse
compress
ion.
20
16.
[13]
Thomps
on,
I.
M.,
et
al.,
Preva
le
n
c
e
of
prostat
e
c
an
ce
r
among
m
en
with
a
prostate
-
spec
if
ic
an
ti
gen
l
eve
l≤
4.
0
ng
per
m
il
li
liter
.
New
E
ngland
Journal
of
Me
d
ic
in
e
,
200
4.
350(22):
p.
22
39
-
2246.
[14]
Hu,
X.,
et
a
l.,
Artifi
c
ial
neur
a
l
ne
tworks
and
prostat
e
c
ancer
—
tools
for
dia
gnosis
and
m
ana
gement.
Nature
Re
views
Ur
ology
,
2013
.
10(3):
p.
174.
[15]
Alexa
ndrid
is,
A.
K.
and
A.D.
Za
p
ran
is,
W
avele
t
n
eur
al
n
et
works
:
with
application
s
in
fina
nc
ial
en
gine
er
ing,
cha
os
,
and
c
la
ss
ifi
c
atio
n.
2014:
John W
il
e
y
&
Sons
.
[16]
Zha
o,
J.
,
W
.
Chen,
and
J.
Luo
.
Fe
ed
forw
ard
wa
ve
l
et
neural
networ
k
and
multi
-
variabl
e
functio
nal
approximation.
in
Int
ern
ationa
l Confere
nc
e
on
C
om
puta
ti
ona
l
an
d
Inform
at
ion
Sc
ie
nc
e. 2004. Spri
nger
.
Evaluation Warning : The document was created with Spire.PDF for Python.